Comparative Studies on Activity Recognition of Elderly People Living Alone

Abstract

The social phenomenon of empty nesters is becoming more and more ubiquitous, and it is difficult to keep a watchful eye on their conditions. The advancement of computer technology and the spread of related applications have promoted the study of pervasive computing and smart cities. Ambient assisted living (AAL) systems have also been born in response to trends and demands. Regarding to medical surveillance, the AAL systems can care for the elderly people living alone full time, provide health advice, and initiate an early warning model when an emergency is detected. Main intention of this paper is to make comparative studies on activity recognition of elderly people living alone utilizing 6 classic classification algorithms, namely decision tree (DT), k-nearest neighbor (KNN), support vector machine (SVM), naive Bayes (NB), linear discriminant analysis (LDA), and ensemble learning (EL). And we adopt these models to recognize 10 activities of daily living (ADLs), namely Meal_Preparation, Relax, Eating, Work, Sleeping, Wash_Dishes, Bed_to_Toilet, Enter_Home, Leave_Home and Housekeeping. In this work, we employ the Aruba annotated open dataset that obtained in a smart house where a voluntary single elderly woman has lived for 220 days. After structuring and cleaning data, we slice data according to entire activity and then extract features from these activity units. Results show that most classification algorithms perform well except for NB based on the activity features we extract. In the future, we can boost performance through improving algorithms, automatically extracting features, and changing the way of reproducing activity representation.